Notably, optimizing these neural networks necessitates a differentiable programming paradigm where gradients of a loss function can be propagated through a numerical solver. In this talk, we will ...